Klarna's CEO told investors in mid-2024 that the company had shut down Salesforce and was weeks away from shutting down Workday. The headline spread everywhere: AI agents are replacing SaaS. The reality, confirmed to CX Today by Klarna itself, was quieter. Klarna had stopped using Workday and Salesforce's CRM-but it replaced them with Deel for HR and a blend of other SaaS tools for CRM functionality. The poster child for "agents kill SaaS" had actually just done a vendor switch.
That gap between the narrative and the facts is worth sitting with.
Why the "AI agents replacing SaaS" debate is framed badly
The claim, as Satya Nadella and others have put it, is that the rise of agentic AI-tools capable of automating and orchestrating tasks across systems-could upend the SaaS model entirely. It is a clean, legible thesis. It is also mostly wrong, in the way that big technology predictions are often wrong: the direction is real, but the mechanism is misidentified.
The thing agents actually threaten is not SaaS as a category. It is a specific kind of SaaS: the point-product tool that exists to give you a UI on top of a database, charges per seat, and quietly prices data extraction at a premium. Some SaaS tools have created "artificial moats" by charging extra for interacting with and extracting data. Now, as more companies move to agentic solutions that interact with this data, they are finding friction-making those tools most at risk for reevaluation.
That is a precise and accurate observation. It is just not the same as "SaaS is dead."
When you look at how enterprises are actually deploying AI agents in 2025 and 2026, they are not replacing their systems of record-they are building orchestration layers on top of them. As one former Microsoft manager noted, a "reality check" is occurring among CIOs as they realize LLMs lack the deterministic consistency required for critical industries like financial services. For use cases such as underwriting, a system that provides a correct answer six out of ten times is insufficient. LLMs interpret human intent; deterministic systems execute the actual work. The deterministic systems are not being disrupted-the operator is.
This is the actual architecture winning in production. Agents sit on top. Systems of record stay.
The consolidation story is more interesting than the replacement story
Here is what the data actually shows. The average enterprise now uses 106 SaaS applications, down from 112 in 2023 and a peak of 130 in 2022-an 18% decline from the peak in just three years. That number had been climbing for a decade. The fact that it is now falling is genuinely significant.
But look at what is driving it. SaaS spending is expected to grow 20% in 2026 even as the number of apps falls. Companies are using fewer tools but paying more for each one-as vendors raise prices, bundle AI features, and move contracts to annual enterprise agreements from month-to-month subscriptions.
The apps are going away. The spend is not. What AI is doing is giving procurement teams a socially acceptable reason to cut the marginal tool they were already embarrassed to be paying for. It is the SaaS equivalent of someone blaming a gym membership cancellation on their "new training philosophy."
The non-obvious consequence: the tools that survive this round of consolidation will be harder to dislodge than ever. If you made the cut, you are now a system of record. If AI can replace the "business logic" layer that SaaS applications provide, then the SaaS layer survives only if it creates unique and indispensable value for customers-and SaaS vendors must innovate, focusing on what AI agents alone cannot do.
Steelman: the replacement thesis is right for one specific category
It is only fair to state the strongest version of the opposing view.
For genuinely thin workflow tools-the kind that exist to move data from one place to another, format a report, or trigger a notification-agents can and do replace the function today. Media publisher 6AM City estimates saving $100,000 a month from replacing its CRM with its own AI-powered version. The agentic development framework Warp is pausing SaaS purchases in favor of agent-based and just-in-time solutions. These are real cases, not thought experiments.
The average enterprise manages 291 SaaS applications-up from just 110 in 2020-and Gartner and Deloitte project that 35% of point-product SaaS tools will be replaced by AI agents or absorbed into agent ecosystems by 2030. A third of the stack going away over five years is a meaningful shift. It is just not the revolution the headline implies.
The pattern that holds across every real deployment: agents replace tasks, not platforms. They replace the human operator clicking through an interface. They do not replace the database, the audit trail, the compliance layer, or the vendor relationship. The difference between SaaS and agents is not autonomy. It is who carries the cognitive burden. A tool requires you to think and operate at its interface. Agents take that burden off you. The tool frequently stays.
Beagle in action#ops-team, Tuesday morningWhat this looks like in practice: the Slack message that used to require someone to open a tool, pull a report, and paste it back gets automated. The tool itself-with its permissions, its audit log, its vendor support contract-is still there. A teammate like Beagle operates at that boundary: it talks to the system of record so a human does not have to context-switch, but the system of record does not go anywhere.
What your team should actually do with this
The right question is not "will agents replace our SaaS stack?" It is "which tools in our stack are charging us to be the interface, rather than to store or process something irreplaceable?"
Three ways to cut through the noise:
- Check for data-extraction fees. If a vendor charges per API call or puts exports behind a higher tier, it has told you exactly how thin its moat is. That is the tool agents are best positioned to route around.
- Separate systems of record from systems of engagement. Your CRM data, your HRIS, your financial ledger-these are not going anywhere. The UI you use to query them is negotiable.
- Watch for the Klarna pattern. Big announcements about abandoning SaaS for AI almost always turn out to be vendor switches in disguise, with a blend of alternative third-party and in-house solutions. That is often the right move. Just be honest about what you are doing.
Agents rely on well-defined objectives, reliable tool APIs, and enough context to make good decisions. The companies winning right now are the ones who understand that constraint and build around it-not the ones treating agents as a drop-in replacement for everything they currently pay for.
The tools that know the difference will do fine. The narrative that does not know the difference will keep misfiring-and the tools priced on that narrative will take the hit they probably deserve.
AI agents replacing SaaS: common questions
Will AI agents actually replace SaaS tools?
Agents will replace specific workflows inside SaaS tools, not the tools themselves. Systems of record-CRMs, HRISs, financial platforms-store data that agents need to query, so they stay. Point-product tools that exist only to provide a UI on top of data access are the real targets, and Gartner projects around 35% of those will be absorbed or replaced by 2030.
What actually happened with Klarna and Salesforce?
Klarna's CEO announced in mid-2024 that the company had shut down Salesforce and would shut down Workday. The reality, confirmed by Klarna, was that they replaced Salesforce's CRM with other tools and adopted Deel as their HR platform. They switched SaaS vendors, not to pure AI. The story became a parable for a thesis it did not actually support.
What kind of SaaS is most at risk from agentic AI?
Tools that charge for data extraction, generate value primarily as a UI layer, or serve a single narrow workflow are most exposed. If an agent can query the underlying data source directly-and the vendor's main value was presenting that data in a formatted interface-the case for the tool weakens quickly.
How should teams think about SaaS consolidation and AI together?
Treat them as separate decisions that often align. Consolidation is already underway for cost and governance reasons; AI gives teams an additional push. The practical move is to identify tools where you are paying for access to your own data, and either renegotiate or route agent workflows around those APIs directly.
Does the "agents on top of SaaS" model actually work in production?
Yes, and it is the dominant pattern in 2025-2026 enterprise deployments. Agents interpret natural-language requests, query connected systems via API or integration, and return sourced answers-leaving the underlying system of record intact. The operator (the human clicking through the interface) is what disappears, not the platform.